AI in Wealth Management: Scaling Advice at Lower Cost

This article distills a recent LinkedIn Live hosted by Marvin Labs. The conversation featured Alex Hoffmann (Co‑Founder & CEO, Marvin Labs), John Goddard (Former COO, HSBC – Global Private Banking and Wealth), and moderator James Yerkess (Former Global Head of Transaction Banking & FX, HSBC Wealth Management).

The panel examined how AI can help banks scale consistent, high‑quality investment advice amid rising cost pressure, where human judgment remains essential, and how falling inference costs from new models like DeepSeek could change adoption economics. Watch the full discussion below.

Cost pressure is forcing operating model change

Goddard framed the discussion with a practical reality: most banks are in active cost‑efficiency and transformation cycles, with headcount the largest cost pool. The near‑term path is to free capacity and reinvest in technology, data, and AI to improve service while lowering the cost‑to‑serve. This mirrors broader industry themes in recent bank reviews that emphasize productivity, operating leverage, and targeted reinvestment in tech and data programs Global Banking Annual Review.

Wealth is lagging retail on AI deployment—for now

The panel agreed that wealth and private banking trail retail banking in applied AI. Banks are running pilots, but most near‑term effort is advisor enablement and back‑to‑front efficiency rather than fully automated client delivery. Competing priorities on legacy platforms and “keep‑the‑lights‑on” spend dilute budgets for net‑new AI builds.

By contrast, fraud and KYC in retail payments have shown visible AI impact. For example, Visa reports its real‑time AI risk models help prevent billions of dollars in attempted fraud annually Visa AI risk solutions.

Standardize the house view, personalize the client conversation

Hoffmann emphasized a pain point in large networks: getting a single, high‑quality investment narrative from central research into thousands of advisor conversations, each tailored to different client segments without diluting the message. Modern AI can synthesize long‑form research and re‑express it for different use cases (portfolio notes, client summaries, institutional briefs) while keeping core guidance consistent.

What you want to avoid is every branch and every relationship manager recreating their own house view. Build it once centrally with the right data and resources, then distribute it cleanly across the organization.

Banks will still want human sign‑off before anything goes directly to end clients, but the standardization and speed gains to the advisor desktop are immediate.

Human in the loop remains essential for trust

Goddard argued that clients still expect a relationship manager to interpret guidance, account for preferences, and help them stay the course during volatile markets. That complements the reality that both humans and algorithms exhibit bias, and clients often forgive human errors more readily than machine errors, a pattern documented as “algorithm aversion” algorithm aversion research.

We are moving to a world where humans and AI work hand in hand. Most clients still want a human involved in big advisory decisions about their future. The question is how to combine the two to get the best result.

Behavioral finance research suggests that individual investors tend to overtrade, which hurts returns trading is hazardous. AI‑supported guardrails and consistent messaging can help reduce that dispersion in outcomes.

Compliance and cross‑border complexity are ripe for AI

Suitability, KYC, documentation, and cross‑border rules are heavy lifts for advisors. The panel highlighted opportunities to embed compliance into the workflow: keep suitability profiles up to date, surface applicable cross‑border rules, and auto‑generate audit trails for supervisory review. Relevant regulatory anchors include the SEC’s Regulation Best Interest SEC Reg BI, ESMA’s suitability guidelines under MiFID II ESMA suitability, and the FCA’s Consumer Duty in the UK FCA Consumer Duty.

Data governance will determine scope and speed

During Q&A, Hoffmann noted Marvin Labs’ private alpha feature that lets users upload broker research to compare areas of consensus and disagreement. Extending into private company data for M&A analysis is feasible from a modeling standpoint, but requires bank‑grade security and willing data partners. Institutions will move faster where they can keep sensitive data in their own tenancy and maintain clear audit trails.

Model economics: falling inference costs change the calculus

The panel touched on how newer models are lowering inference costs and expanding access. Technical reports from DeepSeek highlight training and inference efficiency gains in recent releases DeepSeek‑V3 technical report and emerging reasoning models DeepSeek‑R1. Lower unit costs open more on‑prem or VPC deployment options, which matter for data residency and control in regulated environments. The practical takeaway for banks: re‑run the ROI model for use cases that were marginal in 2023–2024.

What good looks like

• Standardize the narrative centrally and manage quality in one place
• Keep a human accountable at the edge for suitability and client nuance
• Build traceability: sources, versions, and rationale for every recommendation
• Start in lower‑risk domains, measure outcomes, then expand

The best AI is often the one you don’t see. It drafts the report, prepares the tailored brief, and runs risk checks so advisors can focus on the client.

Why this matters for analysts and investors

For coverage analysts, banks’ adoption of AI co‑pilots should compress the research‑to‑advice cycle and produce more consistent messaging across large advisory networks. Expect greater emphasis on validated, data‑backed insights with clear sourcing and audit trails. As inference costs fall and on‑prem options mature, evaluate vendors on security posture, traceability, and integration into primary documents and supervisory workflows.

Watch the full video to hear the complete discussion and perspectives from Alex Hoffmann, John Goddard, and James Yerkess.